Search Baidu web results API

Endpoint facts

Method

GET

API path

/baidu/search

Auth

apiKey

Credits

3 credits/request

Platform

Baidu

Response schema

#/definitions/baidu.searchResponseDoc

MCP tool

baidu_search

Overview

Returns normalized Baidu result cards for a query, read from Baidu's mobile search page: title, destination URL, snippet, publisher, and date, with page-based pagination. Each result's `type` is `web` (standard pages), `video` (video cards), or `baike` (Baidu Baike encyclopedia entries); answer widgets such as weather, entity panels, and recommendation lists are not returned. Baidu challenges a portion of requests with a security verification page; the service retries across browser profiles and returns 503 if every attempt is challenged, so callers should retry on 503. Developers commonly use this endpoint for data enrichment, monitoring, research dashboards, internal automation, and agent-native workflows that need repeatable structured public web data. Authentication uses the documented Crawlora headers, and usage is metered with the credit cost shown on this page.

Authentication

Send your scraping API key in the x-api-key header. Use the console API Keys page to rotate or select the active key.

Billing

Endpoint usage is metered in credits. The plan prices, included credits, limits, and overage rates below match the active backend billing configuration.

Credit cost

3 credits/request

Charged response

Successful 2xx responses

Infrastructure behavior

This endpoint is executed through Crawlora's managed scraping infrastructure.

• Proxy strategy: managed automatically where needed
• Browser rendering: enabled for supported targets that require rendered HTML or JavaScript execution
• Browser cluster: supported dynamic workloads can be routed through distributed browser instances
• Retry behavior: automatic retry/fallback may be used depending on endpoint type
• Challenge handling: challenged pages or unusable upstream HTML are detected and surfaced clearly when they cannot be used
• Billing: credits are charged only for successful 2xx responses
• Observability: responses include request context where available

Browser cluster behavior

Some targets require real browser execution because the data is loaded through JavaScript, dynamic rendering, or interaction-like browser behavior.

For supported endpoints, Crawlora can route requests through a managed browser cluster. This allows Crawlora to execute JavaScript, load dynamic content, apply browser-level request behavior, and normalize the rendered result into JSON.

You do not need to operate your own Playwright, Puppeteer, Chrome, proxy, queue, or retry infrastructure.

Response behavior

- Each result's `type` is one of `web` (standard pages), `video` (video cards), or `baike` (Baidu Baike encyclopedia entries). Answer widgets such as weather panels, entity panels, Tieba cards without a link, and recommendation lists are not results and are not returned. - `source` and `date` are the publisher name and date Baidu shows on the card; `date` is Baidu's own text, such as `2025年10月30日` or `1月19日`, and is omitted when Baidu shows none. - A single page holds about 10 cards, of which roughly 8 to 10 are returned as results. - Baidu challenges a share of requests, and a repeating client address, with a security verification page. The service retries up to three times across mobile browser profiles on fresh connections and never tries to solve the verification; if every attempt is challenged it returns `503`, and the call can simply be retried. - A response page with no result cards and no verification marker is reported as `503` (parser drift), never as an empty success. - `pagination.next_page` is present only when Baidu offers a next page and `page` is below `10`.

Error behavior

Crawlora does not silently return bad data when the upstream page cannot be used.

When possible, Crawlora returns structured error context so your integration can retry, back off, or inspect the request.

Related